[Config Support]: Every detection is gettin label "person"

[bib2]

Describe the problem you are having

Hi, i want to detect many things, including cats, dogs and so on. But every detection is labeld with "person". Even cats are labeld so. With over 90%...

Any idea where i can look at to find the problem?

Version

0.14

Frigate config file

mqtt:
  enabled: false



ffmpeg:
  hwaccel_args: preset-vaapi

# Optional: birdseye configuration
# NOTE: Can (enabled, mode) be overridden at the camera level
birdseye:
  # Optional: Enable birdseye view (default: shown below)
  enabled: true

record:
  enabled: true
  retain:
    days: 3
    mode: motion
  events:
    retain:
      default: 7
      mode: motion

snapshots:
  enabled: true
  retain:
    default: 7

review:
  alerts:
    labels:
      - car
      - cat
      - dog
      - person

cameras:
  eingang_haustuer:
    ffmpeg:
      inputs:
        - path: rtsp://192.168.xxx.xx:xxx/cam/realmonitor?channel=1&subtype=2
          # input_args: preset-rtsp-restream
          roles:
            - detect
        - path: rtsp://192.168.xxx.xx:xxx/cam/realmonitor?channel=1&subtype=0
          roles:
            - record   
    detect:
      width: 1088
      height: 1920
      fps: 5


    motion:
      threshold: 30
      contour_area: 10
      improve_contrast: 'true'
      mask: 0.613,0.039,0.609,0.063,0.963,0.066,0.968,0.041
    zones:
      all:
        coordinates: 0,0,0,1,1,1,1,0
        loitering_time: 0
    objects: {}



    
  sprechanlage:
    ffmpeg:
      inputs:
        - path: rtsp://192.168.xxx.xx:xxx/live/ch00_0
          roles:
            - detect
            - rtmp
    detect:
      width: 1280
      height: 720
      fps: 5
    motion:
      threshold: 30
      contour_area: 10
      improve_contrast: 'true'
    review:
      alerts:
        required_zones:
          - homezone
      detections:
        required_zones:
          - homezone
    zones:
      homezone:
        coordinates: 0.298,0.493,0.986,0.615,0.99,0.991,0.008,0.996,0.004,0.012,0.313,0.007
        loitering_time: 0
detectors:
  ov:
    type: openvino
    device: GPU
model:
  width: 300
  height: 300
  input_tensor: nhwc
  input_pixel_format: bgr
  path: /openvino-model/ssdlite_mobilenet_v2.xml
  labelmap_path: /openvino-model/coco_91cl_bkgr.txt


ui:
  time_format: 24hour
  strftime_fmt: '%d/%m/%Y %H:%M'

version: 0.14
camera_groups:
  Birdseye:
    order: 1
    icon: LuCodepen
    cameras: birdseye

Relevant Frigate log output

...

Relevant go2rtc log output

...

Frigate stats

No response

Operating system

Proxmox

Install method

Proxmox via Docker

docker-compose file or Docker CLI command

proxmox helper file website - standard

Object Detector

CPU (no coral)

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

[halabut69]

SSDLite_Mobilenet prioritises speed over accuracy. Try openvino with the YoloNAS model, to see if it is a model issue.
You can use the notebook here to download the model: https://docs.frigate.video/configuration/object_detectors/#openvino-detector
But due to errors with the download URL, you need to use a workaround to get it to download correctly - see here: #15872

[bib2]

I don´t erxactly understand what to do...

Do i have to go to my docker container and open a cli and enter the pip install command explained in your link to #15872?

Do i have to download the yolo nas manually and put it to the openvino-folder?

[hawkeye217]

You need to add objects to track to the objects section in your config.

objects:
  track:
    - person
    - car
    - ...

https://docs.frigate.video/configuration/objects

[bib2]

I already have this:

 review:
   alerts:
     labels:
       - car
       - cat
       - dog
       - person

Is this not necessary? Should i remove that and only put objects: into my config file?

[hawkeye217]

That only specifies how to classify your detections. The objects --> track config specifies what objects to actually detect. So yes, you need both.

[bib2]

Thank you very much, now it´s the first time, i get a car in my driveway marked as "car".

But i only can see this marks on my android frigate app. Is there a possibility to see the frame with the label name on the webfrontend too? On the webfrontend i only can see the video - but without labels inside the picture.

[hawkeye217]

See the FAQ, Usage question 8.

Frigate 0.15 is in release candidate status and makes this easier.

[bib2]

I want to update my installation to frigate 15, but i don´t know how.

I read a few how-tos on reddit and github, but nothing worked.

I use proxmox, there installed portinaer insade a lxc container. And there inside i installed frigate.

But i don´t know how to update my container to 15.

When i go to my container, i want to edit the url nbut i can´t deploy the new one - field is still grey. I stopped the container, but nthing changed... Am i doing it wrong this way?

I tried it manually, but also i can´t edit the config.yml file of the portainer-docker containwer, don´t know wherei can do this?

[hawkeye217]

None of the maintainers use Proxmox, so it's not an officially supported installation method. Perhaps someone from the community can help.

[bib2]

I´m using a portinaer-container, so i would say that it is installed inside docker?

I can start/stop the docker container with frigate, but i don´t know how to change the url to the new beta-file?

[NickM-27]

it is just changing the image in portainer

[bib2]

But how exactly can i change the image? Do i have to remove the old one in "images" first? Upload a new image inside "images"?

Or do i have to go to "stack" or to "container" section inside portainer?

[NickM-27]

you are editing the container. I can't give specific instructions because I don't use portainer.

This is what gemini says

[bib2]

Yeah, i tried this - i read few how-tos before, but the "deploy" button always stays grey and won´t go blue - so i can´t deploy the new image and i don´t know whats wrong, why it still stays grey.

[NickM-27]

what are you putting for the image

[bib2]

i changed only the last few signs: ***:stable --> *:0.15.0-beta2

But now i went to my promox into the cli of my docker-lxc and i looked for my yml-file and i could find it and change manually, now i´m on 15 beta2.

Thanks.

Will try to set up a new debian VM and install docker inside this by hand instead of lxc containers from helper-scripts.com

[bib2]

But one more question:

Without my objects setting, i got red/orange boxes visible on my recording snapshots labeled with person 91% and so on.

Now with the same values inside object: and inside review: as mentioned above, i get recognitions for all of my objects, car, person, cat and so on. But the boxes are only blue and not red/orange anymore.

So is this normal behaviour or is something misconfigured?

I think red is alert and blue only recognition without alert?

[NickM-27]

do you have an image? I am not sure what part of the UI you are referring to

[bib2]

[NickM-27]

box colors are assigned to objects depending on how many objects they are, the color of the box is not important other than if it is thin dark blue which means it is not detected, or if it is thin grey which means it is stationary

[bib2]

But when i look to my recordings of this morning before all the changes, i only can see red/orange boxes. None of them was blue. And now all is blue only.

[NickM-27]

the color shown in the snapshot depends on the label.

[bib2]

Ok, thanks!

I hope i will get some good recording this night again. Since i used frigate, i didn´t get cat alarm anymore. My old recording system only used movement without AI and i got all cats, hedgehocks, foxes and so on. But since i used frigate, i only got persons... Some cats which were labeled as person...

Hope that now it will work as intended... I have a big problem with animals (cats and even foxes) on my frontdoor, doing their mess on the stairs, carpet and on the door.

[NickM-27]

you will want to make sure it is indeed object detection that is missing, motion detection has its own tuning that needs to be right first

[bib2]

Old system was with knx presence detectors, not with video-recording-software.

When movment was detected, light was powered on and a small videoclip was saved. Will test a few days and learn how to fine tune movement settings inside frigate. Thanks for your patience and help!